Moving averages

• There is a long history of looking at moving averages to try to predict stock returns.
• Moving average = average over a moving window: last 20 days or last 50 days or whatever.
• If a stock is above or below a moving average, or crosses a moving average from above or below, it could be a bullish or bearish sign.

Crude Prices and Moving Averages

Autocorrelation

• Autocorrelation is the correlation of a variable with its own lagged values.
  • At lag 1: does today predict tomorrow? does this month predict next month?
  • At lag 2: does yesterday predict tomorrow? etc.
• Applied to % changes: does today’s % change predict tomorrow’s % change? etc.
  • Positive autocorrelation \(\Rightarrow\) momentum, negative \(\Rightarrow\) reversals

ARMA(1, 1)

• Model is

\[x_t = a + b x_{t-1} + \varepsilon_t + c \varepsilon_{t-1}\]

• \(\varepsilon_t\) are random shocks
• \(b\) is the autoregressive coefficient at lag 1
• \(c\) is the moving average coefficient at lag 1

Autocorrelation and ARMA

Long Short Term Memory

• An LSTM model is a type of recurrent neural network.
• It is widely used for predicting time series data.
• We’ll use the Noteable plugin, because it allows us to install libraries.

Training and Testing an LSTM Model

• The key financial statements are
  • income statement
  • balance sheet
  • statement of cash flows
• The past 5 years are available at finance.yahoo.com.

CVX and COP Financial Statements